xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 8e6ab8837d8d6b9198e67c1c445300b483e2f304)
1 /*	$NetBSD: tcp_subr.c,v 1.144 2003/07/20 16:35:08 he Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1997, 1998, 2000, 2001 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
38  * Facility, NASA Ames Research Center.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. All advertising materials mentioning features or use of this software
49  *    must display the following acknowledgement:
50  *	This product includes software developed by the NetBSD
51  *	Foundation, Inc. and its contributors.
52  * 4. Neither the name of The NetBSD Foundation nor the names of its
53  *    contributors may be used to endorse or promote products derived
54  *    from this software without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
57  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
58  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
59  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
60  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
61  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
62  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
63  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
64  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
65  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
66  * POSSIBILITY OF SUCH DAMAGE.
67  */
68 
69 /*
70  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
71  *	The Regents of the University of California.  All rights reserved.
72  *
73  * Redistribution and use in source and binary forms, with or without
74  * modification, are permitted provided that the following conditions
75  * are met:
76  * 1. Redistributions of source code must retain the above copyright
77  *    notice, this list of conditions and the following disclaimer.
78  * 2. Redistributions in binary form must reproduce the above copyright
79  *    notice, this list of conditions and the following disclaimer in the
80  *    documentation and/or other materials provided with the distribution.
81  * 3. All advertising materials mentioning features or use of this software
82  *    must display the following acknowledgement:
83  *	This product includes software developed by the University of
84  *	California, Berkeley and its contributors.
85  * 4. Neither the name of the University nor the names of its contributors
86  *    may be used to endorse or promote products derived from this software
87  *    without specific prior written permission.
88  *
89  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
90  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
91  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
92  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
93  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
94  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
95  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
96  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
97  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
98  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
99  * SUCH DAMAGE.
100  *
101  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
102  */
103 
104 #include <sys/cdefs.h>
105 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.144 2003/07/20 16:35:08 he Exp $");
106 
107 #include "opt_inet.h"
108 #include "opt_ipsec.h"
109 #include "opt_tcp_compat_42.h"
110 #include "opt_inet_csum.h"
111 #include "opt_mbuftrace.h"
112 #include "rnd.h"
113 
114 #include <sys/param.h>
115 #include <sys/proc.h>
116 #include <sys/systm.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/socket.h>
120 #include <sys/socketvar.h>
121 #include <sys/protosw.h>
122 #include <sys/errno.h>
123 #include <sys/kernel.h>
124 #include <sys/pool.h>
125 #if NRND > 0
126 #include <sys/md5.h>
127 #include <sys/rnd.h>
128 #endif
129 
130 #include <net/route.h>
131 #include <net/if.h>
132 
133 #include <netinet/in.h>
134 #include <netinet/in_systm.h>
135 #include <netinet/ip.h>
136 #include <netinet/in_pcb.h>
137 #include <netinet/ip_var.h>
138 #include <netinet/ip_icmp.h>
139 
140 #ifdef INET6
141 #ifndef INET
142 #include <netinet/in.h>
143 #endif
144 #include <netinet/ip6.h>
145 #include <netinet6/in6_pcb.h>
146 #include <netinet6/ip6_var.h>
147 #include <netinet6/in6_var.h>
148 #include <netinet6/ip6protosw.h>
149 #include <netinet/icmp6.h>
150 #include <netinet6/nd6.h>
151 #endif
152 
153 #include <netinet/tcp.h>
154 #include <netinet/tcp_fsm.h>
155 #include <netinet/tcp_seq.h>
156 #include <netinet/tcp_timer.h>
157 #include <netinet/tcp_var.h>
158 #include <netinet/tcpip.h>
159 
160 #ifdef IPSEC
161 #include <netinet6/ipsec.h>
162 #endif /*IPSEC*/
163 
164 #ifdef INET6
165 struct in6pcb tcb6;
166 #endif
167 
168 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
169 struct	tcpstat tcpstat;	/* tcp statistics */
170 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
171 
172 /* patchable/settable parameters for tcp */
173 int 	tcp_mssdflt = TCP_MSS;
174 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
175 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
176 #if NRND > 0
177 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
178 #endif
179 int	tcp_do_sack = 1;	/* selective acknowledgement */
180 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
181 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
182 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
183 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
184 #ifndef TCP_INIT_WIN
185 #define	TCP_INIT_WIN	1	/* initial slow start window */
186 #endif
187 #ifndef TCP_INIT_WIN_LOCAL
188 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
189 #endif
190 int	tcp_init_win = TCP_INIT_WIN;
191 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
192 int	tcp_mss_ifmtu = 0;
193 #ifdef TCP_COMPAT_42
194 int	tcp_compat_42 = 1;
195 #else
196 int	tcp_compat_42 = 0;
197 #endif
198 int	tcp_rst_ppslim = 100;	/* 100pps */
199 
200 /* tcb hash */
201 #ifndef TCBHASHSIZE
202 #define	TCBHASHSIZE	128
203 #endif
204 int	tcbhashsize = TCBHASHSIZE;
205 
206 /* syn hash parameters */
207 #define	TCP_SYN_HASH_SIZE	293
208 #define	TCP_SYN_BUCKET_SIZE	35
209 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
210 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
211 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
212 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
213 
214 int	tcp_freeq __P((struct tcpcb *));
215 
216 #ifdef INET
217 void	tcp_mtudisc_callback __P((struct in_addr));
218 #endif
219 #ifdef INET6
220 void	tcp6_mtudisc_callback __P((struct in6_addr *));
221 #endif
222 
223 void	tcp_mtudisc __P((struct inpcb *, int));
224 #ifdef INET6
225 void	tcp6_mtudisc __P((struct in6pcb *, int));
226 #endif
227 
228 struct pool tcpcb_pool;
229 
230 #ifdef TCP_CSUM_COUNTERS
231 #include <sys/device.h>
232 
233 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
234     NULL, "tcp", "hwcsum bad");
235 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
236     NULL, "tcp", "hwcsum ok");
237 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
238     NULL, "tcp", "hwcsum data");
239 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
240     NULL, "tcp", "swcsum");
241 #endif /* TCP_CSUM_COUNTERS */
242 
243 #ifdef TCP_OUTPUT_COUNTERS
244 #include <sys/device.h>
245 
246 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
247     NULL, "tcp", "output big header");
248 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
249     NULL, "tcp", "output copy small");
250 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
251     NULL, "tcp", "output copy big");
252 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
253     NULL, "tcp", "output reference big");
254 #endif /* TCP_OUTPUT_COUNTERS */
255 
256 #ifdef TCP_REASS_COUNTERS
257 #include <sys/device.h>
258 
259 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
260     NULL, "tcp_reass", "calls");
261 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
262     &tcp_reass_, "tcp_reass", "insert into empty queue");
263 struct evcnt tcp_reass_iteration[8] = {
264     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
265     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
266     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
267     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
268     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
269     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
270     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
271     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
272 };
273 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
274     &tcp_reass_, "tcp_reass", "prepend to first");
275 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
276     &tcp_reass_, "tcp_reass", "prepend");
277 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278     &tcp_reass_, "tcp_reass", "insert");
279 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280     &tcp_reass_, "tcp_reass", "insert at tail");
281 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282     &tcp_reass_, "tcp_reass", "append");
283 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284     &tcp_reass_, "tcp_reass", "append to tail fragment");
285 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
286     &tcp_reass_, "tcp_reass", "overlap at end");
287 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
288     &tcp_reass_, "tcp_reass", "overlap at start");
289 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
290     &tcp_reass_, "tcp_reass", "duplicate segment");
291 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
292     &tcp_reass_, "tcp_reass", "duplicate fragment");
293 
294 #endif /* TCP_REASS_COUNTERS */
295 
296 #ifdef MBUFTRACE
297 struct mowner tcp_mowner = { "tcp" };
298 struct mowner tcp_rx_mowner = { "tcp", "rx" };
299 struct mowner tcp_tx_mowner = { "tcp", "tx" };
300 #endif
301 
302 /*
303  * Tcp initialization
304  */
305 void
306 tcp_init()
307 {
308 	int hlen;
309 
310 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
311 	    NULL);
312 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
313 #ifdef INET6
314 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
315 #endif
316 
317 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
318 #ifdef INET6
319 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
320 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
321 #endif
322 	if (max_protohdr < hlen)
323 		max_protohdr = hlen;
324 	if (max_linkhdr + hlen > MHLEN)
325 		panic("tcp_init");
326 
327 #ifdef INET
328 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
329 #endif
330 #ifdef INET6
331 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
332 #endif
333 
334 	/* Initialize timer state. */
335 	tcp_timer_init();
336 
337 	/* Initialize the compressed state engine. */
338 	syn_cache_init();
339 
340 #ifdef TCP_CSUM_COUNTERS
341 	evcnt_attach_static(&tcp_hwcsum_bad);
342 	evcnt_attach_static(&tcp_hwcsum_ok);
343 	evcnt_attach_static(&tcp_hwcsum_data);
344 	evcnt_attach_static(&tcp_swcsum);
345 #endif /* TCP_CSUM_COUNTERS */
346 
347 #ifdef TCP_OUTPUT_COUNTERS
348 	evcnt_attach_static(&tcp_output_bigheader);
349 	evcnt_attach_static(&tcp_output_copysmall);
350 	evcnt_attach_static(&tcp_output_copybig);
351 	evcnt_attach_static(&tcp_output_refbig);
352 #endif /* TCP_OUTPUT_COUNTERS */
353 
354 #ifdef TCP_REASS_COUNTERS
355 	evcnt_attach_static(&tcp_reass_);
356 	evcnt_attach_static(&tcp_reass_empty);
357 	evcnt_attach_static(&tcp_reass_iteration[0]);
358 	evcnt_attach_static(&tcp_reass_iteration[1]);
359 	evcnt_attach_static(&tcp_reass_iteration[2]);
360 	evcnt_attach_static(&tcp_reass_iteration[3]);
361 	evcnt_attach_static(&tcp_reass_iteration[4]);
362 	evcnt_attach_static(&tcp_reass_iteration[5]);
363 	evcnt_attach_static(&tcp_reass_iteration[6]);
364 	evcnt_attach_static(&tcp_reass_iteration[7]);
365 	evcnt_attach_static(&tcp_reass_prependfirst);
366 	evcnt_attach_static(&tcp_reass_prepend);
367 	evcnt_attach_static(&tcp_reass_insert);
368 	evcnt_attach_static(&tcp_reass_inserttail);
369 	evcnt_attach_static(&tcp_reass_append);
370 	evcnt_attach_static(&tcp_reass_appendtail);
371 	evcnt_attach_static(&tcp_reass_overlaptail);
372 	evcnt_attach_static(&tcp_reass_overlapfront);
373 	evcnt_attach_static(&tcp_reass_segdup);
374 	evcnt_attach_static(&tcp_reass_fragdup);
375 #endif /* TCP_REASS_COUNTERS */
376 
377 	MOWNER_ATTACH(&tcp_tx_mowner);
378 	MOWNER_ATTACH(&tcp_rx_mowner);
379 	MOWNER_ATTACH(&tcp_mowner);
380 }
381 
382 /*
383  * Create template to be used to send tcp packets on a connection.
384  * Call after host entry created, allocates an mbuf and fills
385  * in a skeletal tcp/ip header, minimizing the amount of work
386  * necessary when the connection is used.
387  */
388 struct mbuf *
389 tcp_template(tp)
390 	struct tcpcb *tp;
391 {
392 	struct inpcb *inp = tp->t_inpcb;
393 #ifdef INET6
394 	struct in6pcb *in6p = tp->t_in6pcb;
395 #endif
396 	struct tcphdr *n;
397 	struct mbuf *m;
398 	int hlen;
399 
400 	switch (tp->t_family) {
401 	case AF_INET:
402 		hlen = sizeof(struct ip);
403 		if (inp)
404 			break;
405 #ifdef INET6
406 		if (in6p) {
407 			/* mapped addr case */
408 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
409 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
410 				break;
411 		}
412 #endif
413 		return NULL;	/*EINVAL*/
414 #ifdef INET6
415 	case AF_INET6:
416 		hlen = sizeof(struct ip6_hdr);
417 		if (in6p) {
418 			/* more sainty check? */
419 			break;
420 		}
421 		return NULL;	/*EINVAL*/
422 #endif
423 	default:
424 		hlen = 0;	/*pacify gcc*/
425 		return NULL;	/*EAFNOSUPPORT*/
426 	}
427 #ifdef DIAGNOSTIC
428 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
429 		panic("mclbytes too small for t_template");
430 #endif
431 	m = tp->t_template;
432 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
433 		;
434 	else {
435 		if (m)
436 			m_freem(m);
437 		m = tp->t_template = NULL;
438 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
439 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
440 			MCLGET(m, M_DONTWAIT);
441 			if ((m->m_flags & M_EXT) == 0) {
442 				m_free(m);
443 				m = NULL;
444 			}
445 		}
446 		if (m == NULL)
447 			return NULL;
448 		MCLAIM(m, &tcp_mowner);
449 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
450 	}
451 
452 	bzero(mtod(m, caddr_t), m->m_len);
453 
454 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
455 
456 	switch (tp->t_family) {
457 	case AF_INET:
458 	    {
459 		struct ipovly *ipov;
460 		mtod(m, struct ip *)->ip_v = 4;
461 		ipov = mtod(m, struct ipovly *);
462 		ipov->ih_pr = IPPROTO_TCP;
463 		ipov->ih_len = htons(sizeof(struct tcphdr));
464 		if (inp) {
465 			ipov->ih_src = inp->inp_laddr;
466 			ipov->ih_dst = inp->inp_faddr;
467 		}
468 #ifdef INET6
469 		else if (in6p) {
470 			/* mapped addr case */
471 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
472 				sizeof(ipov->ih_src));
473 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
474 				sizeof(ipov->ih_dst));
475 		}
476 #endif
477 		/*
478 		 * Compute the pseudo-header portion of the checksum
479 		 * now.  We incrementally add in the TCP option and
480 		 * payload lengths later, and then compute the TCP
481 		 * checksum right before the packet is sent off onto
482 		 * the wire.
483 		 */
484 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
485 		    ipov->ih_dst.s_addr,
486 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
487 		break;
488 	    }
489 #ifdef INET6
490 	case AF_INET6:
491 	    {
492 		struct ip6_hdr *ip6;
493 		mtod(m, struct ip *)->ip_v = 6;
494 		ip6 = mtod(m, struct ip6_hdr *);
495 		ip6->ip6_nxt = IPPROTO_TCP;
496 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
497 		ip6->ip6_src = in6p->in6p_laddr;
498 		ip6->ip6_dst = in6p->in6p_faddr;
499 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
500 		if (ip6_auto_flowlabel) {
501 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
502 			ip6->ip6_flow |=
503 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
504 		}
505 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
506 		ip6->ip6_vfc |= IPV6_VERSION;
507 
508 		/*
509 		 * Compute the pseudo-header portion of the checksum
510 		 * now.  We incrementally add in the TCP option and
511 		 * payload lengths later, and then compute the TCP
512 		 * checksum right before the packet is sent off onto
513 		 * the wire.
514 		 */
515 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
516 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
517 		    htonl(IPPROTO_TCP));
518 		break;
519 	    }
520 #endif
521 	}
522 	if (inp) {
523 		n->th_sport = inp->inp_lport;
524 		n->th_dport = inp->inp_fport;
525 	}
526 #ifdef INET6
527 	else if (in6p) {
528 		n->th_sport = in6p->in6p_lport;
529 		n->th_dport = in6p->in6p_fport;
530 	}
531 #endif
532 	n->th_seq = 0;
533 	n->th_ack = 0;
534 	n->th_x2 = 0;
535 	n->th_off = 5;
536 	n->th_flags = 0;
537 	n->th_win = 0;
538 	n->th_urp = 0;
539 	return (m);
540 }
541 
542 /*
543  * Send a single message to the TCP at address specified by
544  * the given TCP/IP header.  If m == 0, then we make a copy
545  * of the tcpiphdr at ti and send directly to the addressed host.
546  * This is used to force keep alive messages out using the TCP
547  * template for a connection tp->t_template.  If flags are given
548  * then we send a message back to the TCP which originated the
549  * segment ti, and discard the mbuf containing it and any other
550  * attached mbufs.
551  *
552  * In any case the ack and sequence number of the transmitted
553  * segment are as specified by the parameters.
554  */
555 int
556 tcp_respond(tp, template, m, th0, ack, seq, flags)
557 	struct tcpcb *tp;
558 	struct mbuf *template;
559 	struct mbuf *m;
560 	struct tcphdr *th0;
561 	tcp_seq ack, seq;
562 	int flags;
563 {
564 	struct route *ro;
565 	int error, tlen, win = 0;
566 	int hlen;
567 	struct ip *ip;
568 #ifdef INET6
569 	struct ip6_hdr *ip6;
570 #endif
571 	int family;	/* family on packet, not inpcb/in6pcb! */
572 	struct tcphdr *th;
573 
574 	if (tp != NULL && (flags & TH_RST) == 0) {
575 #ifdef DIAGNOSTIC
576 		if (tp->t_inpcb && tp->t_in6pcb)
577 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
578 #endif
579 #ifdef INET
580 		if (tp->t_inpcb)
581 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
582 #endif
583 #ifdef INET6
584 		if (tp->t_in6pcb)
585 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
586 #endif
587 	}
588 
589 	th = NULL;	/* Quell uninitialized warning */
590 	ip = NULL;
591 #ifdef INET6
592 	ip6 = NULL;
593 #endif
594 	if (m == 0) {
595 		if (!template)
596 			return EINVAL;
597 
598 		/* get family information from template */
599 		switch (mtod(template, struct ip *)->ip_v) {
600 		case 4:
601 			family = AF_INET;
602 			hlen = sizeof(struct ip);
603 			break;
604 #ifdef INET6
605 		case 6:
606 			family = AF_INET6;
607 			hlen = sizeof(struct ip6_hdr);
608 			break;
609 #endif
610 		default:
611 			return EAFNOSUPPORT;
612 		}
613 
614 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
615 		if (m) {
616 			MCLAIM(m, &tcp_tx_mowner);
617 			MCLGET(m, M_DONTWAIT);
618 			if ((m->m_flags & M_EXT) == 0) {
619 				m_free(m);
620 				m = NULL;
621 			}
622 		}
623 		if (m == NULL)
624 			return (ENOBUFS);
625 
626 		if (tcp_compat_42)
627 			tlen = 1;
628 		else
629 			tlen = 0;
630 
631 		m->m_data += max_linkhdr;
632 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
633 			template->m_len);
634 		switch (family) {
635 		case AF_INET:
636 			ip = mtod(m, struct ip *);
637 			th = (struct tcphdr *)(ip + 1);
638 			break;
639 #ifdef INET6
640 		case AF_INET6:
641 			ip6 = mtod(m, struct ip6_hdr *);
642 			th = (struct tcphdr *)(ip6 + 1);
643 			break;
644 #endif
645 #if 0
646 		default:
647 			/* noone will visit here */
648 			m_freem(m);
649 			return EAFNOSUPPORT;
650 #endif
651 		}
652 		flags = TH_ACK;
653 	} else {
654 
655 		if ((m->m_flags & M_PKTHDR) == 0) {
656 #if 0
657 			printf("non PKTHDR to tcp_respond\n");
658 #endif
659 			m_freem(m);
660 			return EINVAL;
661 		}
662 #ifdef DIAGNOSTIC
663 		if (!th0)
664 			panic("th0 == NULL in tcp_respond");
665 #endif
666 
667 		/* get family information from m */
668 		switch (mtod(m, struct ip *)->ip_v) {
669 		case 4:
670 			family = AF_INET;
671 			hlen = sizeof(struct ip);
672 			ip = mtod(m, struct ip *);
673 			break;
674 #ifdef INET6
675 		case 6:
676 			family = AF_INET6;
677 			hlen = sizeof(struct ip6_hdr);
678 			ip6 = mtod(m, struct ip6_hdr *);
679 			break;
680 #endif
681 		default:
682 			m_freem(m);
683 			return EAFNOSUPPORT;
684 		}
685 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
686 			tlen = sizeof(*th0);
687 		else
688 			tlen = th0->th_off << 2;
689 
690 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
691 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
692 			m->m_len = hlen + tlen;
693 			m_freem(m->m_next);
694 			m->m_next = NULL;
695 		} else {
696 			struct mbuf *n;
697 
698 #ifdef DIAGNOSTIC
699 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
700 				m_freem(m);
701 				return EMSGSIZE;
702 			}
703 #endif
704 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
705 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
706 				MCLGET(n, M_DONTWAIT);
707 				if ((n->m_flags & M_EXT) == 0) {
708 					m_freem(n);
709 					n = NULL;
710 				}
711 			}
712 			if (!n) {
713 				m_freem(m);
714 				return ENOBUFS;
715 			}
716 
717 			MCLAIM(n, &tcp_tx_mowner);
718 			n->m_data += max_linkhdr;
719 			n->m_len = hlen + tlen;
720 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
721 			m_copyback(n, hlen, tlen, (caddr_t)th0);
722 
723 			m_freem(m);
724 			m = n;
725 			n = NULL;
726 		}
727 
728 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
729 		switch (family) {
730 		case AF_INET:
731 			ip = mtod(m, struct ip *);
732 			th = (struct tcphdr *)(ip + 1);
733 			ip->ip_p = IPPROTO_TCP;
734 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
735 			ip->ip_p = IPPROTO_TCP;
736 			break;
737 #ifdef INET6
738 		case AF_INET6:
739 			ip6 = mtod(m, struct ip6_hdr *);
740 			th = (struct tcphdr *)(ip6 + 1);
741 			ip6->ip6_nxt = IPPROTO_TCP;
742 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
743 			ip6->ip6_nxt = IPPROTO_TCP;
744 			break;
745 #endif
746 #if 0
747 		default:
748 			/* noone will visit here */
749 			m_freem(m);
750 			return EAFNOSUPPORT;
751 #endif
752 		}
753 		xchg(th->th_dport, th->th_sport, u_int16_t);
754 #undef xchg
755 		tlen = 0;	/*be friendly with the following code*/
756 	}
757 	th->th_seq = htonl(seq);
758 	th->th_ack = htonl(ack);
759 	th->th_x2 = 0;
760 	if ((flags & TH_SYN) == 0) {
761 		if (tp)
762 			win >>= tp->rcv_scale;
763 		if (win > TCP_MAXWIN)
764 			win = TCP_MAXWIN;
765 		th->th_win = htons((u_int16_t)win);
766 		th->th_off = sizeof (struct tcphdr) >> 2;
767 		tlen += sizeof(*th);
768 	} else
769 		tlen += th->th_off << 2;
770 	m->m_len = hlen + tlen;
771 	m->m_pkthdr.len = hlen + tlen;
772 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
773 	th->th_flags = flags;
774 	th->th_urp = 0;
775 
776 	switch (family) {
777 #ifdef INET
778 	case AF_INET:
779 	    {
780 		struct ipovly *ipov = (struct ipovly *)ip;
781 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
782 		ipov->ih_len = htons((u_int16_t)tlen);
783 
784 		th->th_sum = 0;
785 		th->th_sum = in_cksum(m, hlen + tlen);
786 		ip->ip_len = htons(hlen + tlen);
787 		ip->ip_ttl = ip_defttl;
788 		break;
789 	    }
790 #endif
791 #ifdef INET6
792 	case AF_INET6:
793 	    {
794 		th->th_sum = 0;
795 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
796 				tlen);
797 		ip6->ip6_plen = ntohs(tlen);
798 		if (tp && tp->t_in6pcb) {
799 			struct ifnet *oifp;
800 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
801 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
802 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
803 		} else
804 			ip6->ip6_hlim = ip6_defhlim;
805 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
806 		if (ip6_auto_flowlabel) {
807 			ip6->ip6_flow |=
808 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
809 		}
810 		break;
811 	    }
812 #endif
813 	}
814 
815 #ifdef IPSEC
816 	(void)ipsec_setsocket(m, NULL);
817 #endif /*IPSEC*/
818 
819 	if (tp != NULL && tp->t_inpcb != NULL) {
820 		ro = &tp->t_inpcb->inp_route;
821 #ifdef IPSEC
822 		if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) {
823 			m_freem(m);
824 			return ENOBUFS;
825 		}
826 #endif
827 #ifdef DIAGNOSTIC
828 		if (family != AF_INET)
829 			panic("tcp_respond: address family mismatch");
830 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
831 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
832 			    ntohl(ip->ip_dst.s_addr),
833 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
834 		}
835 #endif
836 	}
837 #ifdef INET6
838 	else if (tp != NULL && tp->t_in6pcb != NULL) {
839 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
840 #ifdef IPSEC
841 		if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) {
842 			m_freem(m);
843 			return ENOBUFS;
844 		}
845 #endif
846 #ifdef DIAGNOSTIC
847 		if (family == AF_INET) {
848 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
849 				panic("tcp_respond: not mapped addr");
850 			if (bcmp(&ip->ip_dst,
851 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
852 			    sizeof(ip->ip_dst)) != 0) {
853 				panic("tcp_respond: ip_dst != in6p_faddr");
854 			}
855 		} else if (family == AF_INET6) {
856 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
857 			    &tp->t_in6pcb->in6p_faddr))
858 				panic("tcp_respond: ip6_dst != in6p_faddr");
859 		} else
860 			panic("tcp_respond: address family mismatch");
861 #endif
862 	}
863 #endif
864 	else
865 		ro = NULL;
866 
867 	switch (family) {
868 #ifdef INET
869 	case AF_INET:
870 		error = ip_output(m, NULL, ro,
871 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
872 		    NULL);
873 		break;
874 #endif
875 #ifdef INET6
876 	case AF_INET6:
877 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL,
878 		    NULL);
879 		break;
880 #endif
881 	default:
882 		error = EAFNOSUPPORT;
883 		break;
884 	}
885 
886 	return (error);
887 }
888 
889 /*
890  * Create a new TCP control block, making an
891  * empty reassembly queue and hooking it to the argument
892  * protocol control block.
893  */
894 struct tcpcb *
895 tcp_newtcpcb(family, aux)
896 	int family;	/* selects inpcb, or in6pcb */
897 	void *aux;
898 {
899 	struct tcpcb *tp;
900 	int i;
901 
902 	switch (family) {
903 	case PF_INET:
904 		break;
905 #ifdef INET6
906 	case PF_INET6:
907 		break;
908 #endif
909 	default:
910 		return NULL;
911 	}
912 
913 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
914 	if (tp == NULL)
915 		return (NULL);
916 	bzero((caddr_t)tp, sizeof(struct tcpcb));
917 	TAILQ_INIT(&tp->segq);
918 	TAILQ_INIT(&tp->timeq);
919 	tp->t_family = family;		/* may be overridden later on */
920 	tp->t_peermss = tcp_mssdflt;
921 	tp->t_ourmss = tcp_mssdflt;
922 	tp->t_segsz = tcp_mssdflt;
923 	LIST_INIT(&tp->t_sc);
924 
925 	tp->t_lastm = NULL;
926 	tp->t_lastoff = 0;
927 
928 	callout_init(&tp->t_delack_ch);
929 	for (i = 0; i < TCPT_NTIMERS; i++)
930 		TCP_TIMER_INIT(tp, i);
931 
932 	tp->t_flags = 0;
933 	if (tcp_do_rfc1323 && tcp_do_win_scale)
934 		tp->t_flags |= TF_REQ_SCALE;
935 	if (tcp_do_rfc1323 && tcp_do_timestamps)
936 		tp->t_flags |= TF_REQ_TSTMP;
937 	if (tcp_do_sack == 2)
938 		tp->t_flags |= TF_WILL_SACK;
939 	else if (tcp_do_sack == 1)
940 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
941 	tp->t_flags |= TF_CANT_TXSACK;
942 	switch (family) {
943 	case PF_INET:
944 		tp->t_inpcb = (struct inpcb *)aux;
945 		tp->t_mtudisc = ip_mtudisc;
946 		break;
947 #ifdef INET6
948 	case PF_INET6:
949 		tp->t_in6pcb = (struct in6pcb *)aux;
950 		/* for IPv6, always try to run path MTU discovery */
951 		tp->t_mtudisc = 1;
952 		break;
953 #endif
954 	}
955 	/*
956 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
957 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
958 	 * reasonable initial retransmit time.
959 	 */
960 	tp->t_srtt = TCPTV_SRTTBASE;
961 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
962 	tp->t_rttmin = TCPTV_MIN;
963 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
964 	    TCPTV_MIN, TCPTV_REXMTMAX);
965 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
966 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
967 	if (family == AF_INET) {
968 		struct inpcb *inp = (struct inpcb *)aux;
969 		inp->inp_ip.ip_ttl = ip_defttl;
970 		inp->inp_ppcb = (caddr_t)tp;
971 	}
972 #ifdef INET6
973 	else if (family == AF_INET6) {
974 		struct in6pcb *in6p = (struct in6pcb *)aux;
975 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
976 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
977 					       : NULL);
978 		in6p->in6p_ppcb = (caddr_t)tp;
979 	}
980 #endif
981 
982 	/*
983 	 * Initialize our timebase.  When we send timestamps, we take
984 	 * the delta from tcp_now -- this means each connection always
985 	 * gets a timebase of 0, which makes it, among other things,
986 	 * more difficult to determine how long a system has been up,
987 	 * and thus how many TCP sequence increments have occurred.
988 	 */
989 	tp->ts_timebase = tcp_now;
990 
991 	return (tp);
992 }
993 
994 /*
995  * Drop a TCP connection, reporting
996  * the specified error.  If connection is synchronized,
997  * then send a RST to peer.
998  */
999 struct tcpcb *
1000 tcp_drop(tp, errno)
1001 	struct tcpcb *tp;
1002 	int errno;
1003 {
1004 	struct socket *so = NULL;
1005 
1006 #ifdef DIAGNOSTIC
1007 	if (tp->t_inpcb && tp->t_in6pcb)
1008 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1009 #endif
1010 #ifdef INET
1011 	if (tp->t_inpcb)
1012 		so = tp->t_inpcb->inp_socket;
1013 #endif
1014 #ifdef INET6
1015 	if (tp->t_in6pcb)
1016 		so = tp->t_in6pcb->in6p_socket;
1017 #endif
1018 	if (!so)
1019 		return NULL;
1020 
1021 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1022 		tp->t_state = TCPS_CLOSED;
1023 		(void) tcp_output(tp);
1024 		tcpstat.tcps_drops++;
1025 	} else
1026 		tcpstat.tcps_conndrops++;
1027 	if (errno == ETIMEDOUT && tp->t_softerror)
1028 		errno = tp->t_softerror;
1029 	so->so_error = errno;
1030 	return (tcp_close(tp));
1031 }
1032 
1033 /*
1034  * Return whether this tcpcb is marked as dead, indicating
1035  * to the calling timer function that no further action should
1036  * be taken, as we are about to release this tcpcb.  The release
1037  * of the storage will be done if this is the last timer running.
1038  *
1039  * This is typically called from the callout handler function before
1040  * callout_ack() is done, therefore we need to test the number of
1041  * running timer functions against 1 below, not 0.
1042  */
1043 int
1044 tcp_isdead(tp)
1045 	struct tcpcb *tp;
1046 {
1047 	int dead = (tp->t_flags & TF_DEAD);
1048 
1049 	if (__predict_false(dead)) {
1050 		if (tcp_timers_invoking(tp) > 1)
1051 				/* not quite there yet -- count separately? */
1052 			return dead;
1053 		tcpstat.tcps_delayed_free++;
1054 		pool_put(&tcpcb_pool, tp);
1055 	}
1056 	return dead;
1057 }
1058 
1059 /*
1060  * Close a TCP control block:
1061  *	discard all space held by the tcp
1062  *	discard internet protocol block
1063  *	wake up any sleepers
1064  */
1065 struct tcpcb *
1066 tcp_close(tp)
1067 	struct tcpcb *tp;
1068 {
1069 	struct inpcb *inp;
1070 #ifdef INET6
1071 	struct in6pcb *in6p;
1072 #endif
1073 	struct socket *so;
1074 #ifdef RTV_RTT
1075 	struct rtentry *rt;
1076 #endif
1077 	struct route *ro;
1078 
1079 	inp = tp->t_inpcb;
1080 #ifdef INET6
1081 	in6p = tp->t_in6pcb;
1082 #endif
1083 	so = NULL;
1084 	ro = NULL;
1085 	if (inp) {
1086 		so = inp->inp_socket;
1087 		ro = &inp->inp_route;
1088 	}
1089 #ifdef INET6
1090 	else if (in6p) {
1091 		so = in6p->in6p_socket;
1092 		ro = (struct route *)&in6p->in6p_route;
1093 	}
1094 #endif
1095 
1096 #ifdef RTV_RTT
1097 	/*
1098 	 * If we sent enough data to get some meaningful characteristics,
1099 	 * save them in the routing entry.  'Enough' is arbitrarily
1100 	 * defined as the sendpipesize (default 4K) * 16.  This would
1101 	 * give us 16 rtt samples assuming we only get one sample per
1102 	 * window (the usual case on a long haul net).  16 samples is
1103 	 * enough for the srtt filter to converge to within 5% of the correct
1104 	 * value; fewer samples and we could save a very bogus rtt.
1105 	 *
1106 	 * Don't update the default route's characteristics and don't
1107 	 * update anything that the user "locked".
1108 	 */
1109 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1110 	    ro && (rt = ro->ro_rt) &&
1111 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1112 		u_long i = 0;
1113 
1114 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1115 			i = tp->t_srtt *
1116 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1117 			if (rt->rt_rmx.rmx_rtt && i)
1118 				/*
1119 				 * filter this update to half the old & half
1120 				 * the new values, converting scale.
1121 				 * See route.h and tcp_var.h for a
1122 				 * description of the scaling constants.
1123 				 */
1124 				rt->rt_rmx.rmx_rtt =
1125 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1126 			else
1127 				rt->rt_rmx.rmx_rtt = i;
1128 		}
1129 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1130 			i = tp->t_rttvar *
1131 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1132 			if (rt->rt_rmx.rmx_rttvar && i)
1133 				rt->rt_rmx.rmx_rttvar =
1134 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1135 			else
1136 				rt->rt_rmx.rmx_rttvar = i;
1137 		}
1138 		/*
1139 		 * update the pipelimit (ssthresh) if it has been updated
1140 		 * already or if a pipesize was specified & the threshhold
1141 		 * got below half the pipesize.  I.e., wait for bad news
1142 		 * before we start updating, then update on both good
1143 		 * and bad news.
1144 		 */
1145 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1146 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1147 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1148 			/*
1149 			 * convert the limit from user data bytes to
1150 			 * packets then to packet data bytes.
1151 			 */
1152 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1153 			if (i < 2)
1154 				i = 2;
1155 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1156 			if (rt->rt_rmx.rmx_ssthresh)
1157 				rt->rt_rmx.rmx_ssthresh =
1158 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1159 			else
1160 				rt->rt_rmx.rmx_ssthresh = i;
1161 		}
1162 	}
1163 #endif /* RTV_RTT */
1164 	/* free the reassembly queue, if any */
1165 	TCP_REASS_LOCK(tp);
1166 	(void) tcp_freeq(tp);
1167 	TCP_REASS_UNLOCK(tp);
1168 
1169 	tcp_canceltimers(tp);
1170 	TCP_CLEAR_DELACK(tp);
1171 	syn_cache_cleanup(tp);
1172 
1173 	if (tp->t_template) {
1174 		m_free(tp->t_template);
1175 		tp->t_template = NULL;
1176 	}
1177 	if (tcp_timers_invoking(tp))
1178 		tp->t_flags |= TF_DEAD;
1179 	else
1180 		pool_put(&tcpcb_pool, tp);
1181 
1182 	if (inp) {
1183 		inp->inp_ppcb = 0;
1184 		soisdisconnected(so);
1185 		in_pcbdetach(inp);
1186 	}
1187 #ifdef INET6
1188 	else if (in6p) {
1189 		in6p->in6p_ppcb = 0;
1190 		soisdisconnected(so);
1191 		in6_pcbdetach(in6p);
1192 	}
1193 #endif
1194 	tcpstat.tcps_closed++;
1195 	return ((struct tcpcb *)0);
1196 }
1197 
1198 int
1199 tcp_freeq(tp)
1200 	struct tcpcb *tp;
1201 {
1202 	struct ipqent *qe;
1203 	int rv = 0;
1204 #ifdef TCPREASS_DEBUG
1205 	int i = 0;
1206 #endif
1207 
1208 	TCP_REASS_LOCK_CHECK(tp);
1209 
1210 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1211 #ifdef TCPREASS_DEBUG
1212 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1213 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1214 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1215 #endif
1216 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1217 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1218 		m_freem(qe->ipqe_m);
1219 		pool_put(&ipqent_pool, qe);
1220 		rv = 1;
1221 	}
1222 	return (rv);
1223 }
1224 
1225 /*
1226  * Protocol drain routine.  Called when memory is in short supply.
1227  */
1228 void
1229 tcp_drain()
1230 {
1231 	struct inpcb *inp;
1232 	struct tcpcb *tp;
1233 
1234 	/*
1235 	 * Free the sequence queue of all TCP connections.
1236 	 */
1237 	inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue);
1238 	if (inp)						/* XXX */
1239 	CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
1240 		if ((tp = intotcpcb(inp)) != NULL) {
1241 			/*
1242 			 * We may be called from a device's interrupt
1243 			 * context.  If the tcpcb is already busy,
1244 			 * just bail out now.
1245 			 */
1246 			if (tcp_reass_lock_try(tp) == 0)
1247 				continue;
1248 			if (tcp_freeq(tp))
1249 				tcpstat.tcps_connsdrained++;
1250 			TCP_REASS_UNLOCK(tp);
1251 		}
1252 	}
1253 }
1254 
1255 #ifdef INET6
1256 void
1257 tcp6_drain()
1258 {
1259 	struct in6pcb *in6p;
1260 	struct tcpcb *tp;
1261 	struct in6pcb *head = &tcb6;
1262 
1263 	/*
1264 	 * Free the sequence queue of all TCP connections.
1265 	 */
1266 	for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) {
1267 		if ((tp = in6totcpcb(in6p)) != NULL) {
1268 			/*
1269 			 * We may be called from a device's interrupt
1270 			 * context.  If the tcpcb is already busy,
1271 			 * just bail out now.
1272 			 */
1273 			if (tcp_reass_lock_try(tp) == 0)
1274 				continue;
1275 			if (tcp_freeq(tp))
1276 				tcpstat.tcps_connsdrained++;
1277 			TCP_REASS_UNLOCK(tp);
1278 		}
1279 	}
1280 }
1281 #endif
1282 
1283 /*
1284  * Notify a tcp user of an asynchronous error;
1285  * store error as soft error, but wake up user
1286  * (for now, won't do anything until can select for soft error).
1287  */
1288 void
1289 tcp_notify(inp, error)
1290 	struct inpcb *inp;
1291 	int error;
1292 {
1293 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1294 	struct socket *so = inp->inp_socket;
1295 
1296 	/*
1297 	 * Ignore some errors if we are hooked up.
1298 	 * If connection hasn't completed, has retransmitted several times,
1299 	 * and receives a second error, give up now.  This is better
1300 	 * than waiting a long time to establish a connection that
1301 	 * can never complete.
1302 	 */
1303 	if (tp->t_state == TCPS_ESTABLISHED &&
1304 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1305 	      error == EHOSTDOWN)) {
1306 		return;
1307 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1308 	    tp->t_rxtshift > 3 && tp->t_softerror)
1309 		so->so_error = error;
1310 	else
1311 		tp->t_softerror = error;
1312 	wakeup((caddr_t) &so->so_timeo);
1313 	sorwakeup(so);
1314 	sowwakeup(so);
1315 }
1316 
1317 #ifdef INET6
1318 void
1319 tcp6_notify(in6p, error)
1320 	struct in6pcb *in6p;
1321 	int error;
1322 {
1323 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1324 	struct socket *so = in6p->in6p_socket;
1325 
1326 	/*
1327 	 * Ignore some errors if we are hooked up.
1328 	 * If connection hasn't completed, has retransmitted several times,
1329 	 * and receives a second error, give up now.  This is better
1330 	 * than waiting a long time to establish a connection that
1331 	 * can never complete.
1332 	 */
1333 	if (tp->t_state == TCPS_ESTABLISHED &&
1334 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1335 	      error == EHOSTDOWN)) {
1336 		return;
1337 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1338 	    tp->t_rxtshift > 3 && tp->t_softerror)
1339 		so->so_error = error;
1340 	else
1341 		tp->t_softerror = error;
1342 	wakeup((caddr_t) &so->so_timeo);
1343 	sorwakeup(so);
1344 	sowwakeup(so);
1345 }
1346 #endif
1347 
1348 #ifdef INET6
1349 void
1350 tcp6_ctlinput(cmd, sa, d)
1351 	int cmd;
1352 	struct sockaddr *sa;
1353 	void *d;
1354 {
1355 	struct tcphdr th;
1356 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1357 	int nmatch;
1358 	struct ip6_hdr *ip6;
1359 	const struct sockaddr_in6 *sa6_src = NULL;
1360 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1361 	struct mbuf *m;
1362 	int off;
1363 
1364 	if (sa->sa_family != AF_INET6 ||
1365 	    sa->sa_len != sizeof(struct sockaddr_in6))
1366 		return;
1367 	if ((unsigned)cmd >= PRC_NCMDS)
1368 		return;
1369 	else if (cmd == PRC_QUENCH) {
1370 		/* XXX there's no PRC_QUENCH in IPv6 */
1371 		notify = tcp6_quench;
1372 	} else if (PRC_IS_REDIRECT(cmd))
1373 		notify = in6_rtchange, d = NULL;
1374 	else if (cmd == PRC_MSGSIZE)
1375 		; /* special code is present, see below */
1376 	else if (cmd == PRC_HOSTDEAD)
1377 		d = NULL;
1378 	else if (inet6ctlerrmap[cmd] == 0)
1379 		return;
1380 
1381 	/* if the parameter is from icmp6, decode it. */
1382 	if (d != NULL) {
1383 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1384 		m = ip6cp->ip6c_m;
1385 		ip6 = ip6cp->ip6c_ip6;
1386 		off = ip6cp->ip6c_off;
1387 		sa6_src = ip6cp->ip6c_src;
1388 	} else {
1389 		m = NULL;
1390 		ip6 = NULL;
1391 		sa6_src = &sa6_any;
1392 	}
1393 
1394 	if (ip6) {
1395 		/*
1396 		 * XXX: We assume that when ip6 is non NULL,
1397 		 * M and OFF are valid.
1398 		 */
1399 
1400 		/* check if we can safely examine src and dst ports */
1401 		if (m->m_pkthdr.len < off + sizeof(th)) {
1402 			if (cmd == PRC_MSGSIZE)
1403 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1404 			return;
1405 		}
1406 
1407 		bzero(&th, sizeof(th));
1408 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1409 
1410 		if (cmd == PRC_MSGSIZE) {
1411 			int valid = 0;
1412 
1413 			/*
1414 			 * Check to see if we have a valid TCP connection
1415 			 * corresponding to the address in the ICMPv6 message
1416 			 * payload.
1417 			 */
1418 			if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr,
1419 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1420 			    th.th_sport, 0))
1421 				valid++;
1422 
1423 			/*
1424 			 * Depending on the value of "valid" and routing table
1425 			 * size (mtudisc_{hi,lo}wat), we will:
1426 			 * - recalcurate the new MTU and create the
1427 			 *   corresponding routing entry, or
1428 			 * - ignore the MTU change notification.
1429 			 */
1430 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1431 
1432 			/*
1433 			 * no need to call in6_pcbnotify, it should have been
1434 			 * called via callback if necessary
1435 			 */
1436 			return;
1437 		}
1438 
1439 		nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport,
1440 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1441 		if (nmatch == 0 && syn_cache_count &&
1442 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1443 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1444 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1445 			syn_cache_unreach((struct sockaddr *)sa6_src,
1446 					  sa, &th);
1447 	} else {
1448 		(void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src,
1449 		    0, cmd, NULL, notify);
1450 	}
1451 }
1452 #endif
1453 
1454 #ifdef INET
1455 /* assumes that ip header and tcp header are contiguous on mbuf */
1456 void *
1457 tcp_ctlinput(cmd, sa, v)
1458 	int cmd;
1459 	struct sockaddr *sa;
1460 	void *v;
1461 {
1462 	struct ip *ip = v;
1463 	struct tcphdr *th;
1464 	struct icmp *icp;
1465 	extern const int inetctlerrmap[];
1466 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1467 	int errno;
1468 	int nmatch;
1469 #ifdef INET6
1470 	struct in6_addr src6, dst6;
1471 #endif
1472 
1473 	if (sa->sa_family != AF_INET ||
1474 	    sa->sa_len != sizeof(struct sockaddr_in))
1475 		return NULL;
1476 	if ((unsigned)cmd >= PRC_NCMDS)
1477 		return NULL;
1478 	errno = inetctlerrmap[cmd];
1479 	if (cmd == PRC_QUENCH)
1480 		notify = tcp_quench;
1481 	else if (PRC_IS_REDIRECT(cmd))
1482 		notify = in_rtchange, ip = 0;
1483 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1484 		/*
1485 		 * Check to see if we have a valid TCP connection
1486 		 * corresponding to the address in the ICMP message
1487 		 * payload.
1488 		 *
1489 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1490 		 */
1491 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1492 #ifdef INET6
1493 		memset(&src6, 0, sizeof(src6));
1494 		memset(&dst6, 0, sizeof(dst6));
1495 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1496 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1497 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1498 #endif
1499 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1500 		    ip->ip_src, th->th_sport) != NULL)
1501 			;
1502 #ifdef INET6
1503 		else if (in6_pcblookup_connect(&tcb6, &dst6,
1504 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1505 			;
1506 #endif
1507 		else
1508 			return NULL;
1509 
1510 		/*
1511 		 * Now that we've validated that we are actually communicating
1512 		 * with the host indicated in the ICMP message, locate the
1513 		 * ICMP header, recalculate the new MTU, and create the
1514 		 * corresponding routing entry.
1515 		 */
1516 		icp = (struct icmp *)((caddr_t)ip -
1517 		    offsetof(struct icmp, icmp_ip));
1518 		icmp_mtudisc(icp, ip->ip_dst);
1519 
1520 		return NULL;
1521 	} else if (cmd == PRC_HOSTDEAD)
1522 		ip = 0;
1523 	else if (errno == 0)
1524 		return NULL;
1525 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1526 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1527 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1528 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1529 		if (nmatch == 0 && syn_cache_count &&
1530 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1531 		    inetctlerrmap[cmd] == ENETUNREACH ||
1532 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1533 			struct sockaddr_in sin;
1534 			bzero(&sin, sizeof(sin));
1535 			sin.sin_len = sizeof(sin);
1536 			sin.sin_family = AF_INET;
1537 			sin.sin_port = th->th_sport;
1538 			sin.sin_addr = ip->ip_src;
1539 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1540 		}
1541 
1542 		/* XXX mapped address case */
1543 	} else
1544 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1545 		    notify);
1546 	return NULL;
1547 }
1548 
1549 /*
1550  * When a source quence is received, we are being notifed of congestion.
1551  * Close the congestion window down to the Loss Window (one segment).
1552  * We will gradually open it again as we proceed.
1553  */
1554 void
1555 tcp_quench(inp, errno)
1556 	struct inpcb *inp;
1557 	int errno;
1558 {
1559 	struct tcpcb *tp = intotcpcb(inp);
1560 
1561 	if (tp)
1562 		tp->snd_cwnd = tp->t_segsz;
1563 }
1564 #endif
1565 
1566 #ifdef INET6
1567 void
1568 tcp6_quench(in6p, errno)
1569 	struct in6pcb *in6p;
1570 	int errno;
1571 {
1572 	struct tcpcb *tp = in6totcpcb(in6p);
1573 
1574 	if (tp)
1575 		tp->snd_cwnd = tp->t_segsz;
1576 }
1577 #endif
1578 
1579 #ifdef INET
1580 /*
1581  * Path MTU Discovery handlers.
1582  */
1583 void
1584 tcp_mtudisc_callback(faddr)
1585 	struct in_addr faddr;
1586 {
1587 #ifdef INET6
1588 	struct in6_addr in6;
1589 #endif
1590 
1591 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1592 #ifdef INET6
1593 	memset(&in6, 0, sizeof(in6));
1594 	in6.s6_addr16[5] = 0xffff;
1595 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1596 	tcp6_mtudisc_callback(&in6);
1597 #endif
1598 }
1599 
1600 /*
1601  * On receipt of path MTU corrections, flush old route and replace it
1602  * with the new one.  Retransmit all unacknowledged packets, to ensure
1603  * that all packets will be received.
1604  */
1605 void
1606 tcp_mtudisc(inp, errno)
1607 	struct inpcb *inp;
1608 	int errno;
1609 {
1610 	struct tcpcb *tp = intotcpcb(inp);
1611 	struct rtentry *rt = in_pcbrtentry(inp);
1612 
1613 	if (tp != 0) {
1614 		if (rt != 0) {
1615 			/*
1616 			 * If this was not a host route, remove and realloc.
1617 			 */
1618 			if ((rt->rt_flags & RTF_HOST) == 0) {
1619 				in_rtchange(inp, errno);
1620 				if ((rt = in_pcbrtentry(inp)) == 0)
1621 					return;
1622 			}
1623 
1624 			/*
1625 			 * Slow start out of the error condition.  We
1626 			 * use the MTU because we know it's smaller
1627 			 * than the previously transmitted segment.
1628 			 *
1629 			 * Note: This is more conservative than the
1630 			 * suggestion in draft-floyd-incr-init-win-03.
1631 			 */
1632 			if (rt->rt_rmx.rmx_mtu != 0)
1633 				tp->snd_cwnd =
1634 				    TCP_INITIAL_WINDOW(tcp_init_win,
1635 				    rt->rt_rmx.rmx_mtu);
1636 		}
1637 
1638 		/*
1639 		 * Resend unacknowledged packets.
1640 		 */
1641 		tp->snd_nxt = tp->snd_una;
1642 		tcp_output(tp);
1643 	}
1644 }
1645 #endif
1646 
1647 #ifdef INET6
1648 /*
1649  * Path MTU Discovery handlers.
1650  */
1651 void
1652 tcp6_mtudisc_callback(faddr)
1653 	struct in6_addr *faddr;
1654 {
1655 	struct sockaddr_in6 sin6;
1656 
1657 	bzero(&sin6, sizeof(sin6));
1658 	sin6.sin6_family = AF_INET6;
1659 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1660 	sin6.sin6_addr = *faddr;
1661 	(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0,
1662 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1663 }
1664 
1665 void
1666 tcp6_mtudisc(in6p, errno)
1667 	struct in6pcb *in6p;
1668 	int errno;
1669 {
1670 	struct tcpcb *tp = in6totcpcb(in6p);
1671 	struct rtentry *rt = in6_pcbrtentry(in6p);
1672 
1673 	if (tp != 0) {
1674 		if (rt != 0) {
1675 			/*
1676 			 * If this was not a host route, remove and realloc.
1677 			 */
1678 			if ((rt->rt_flags & RTF_HOST) == 0) {
1679 				in6_rtchange(in6p, errno);
1680 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1681 					return;
1682 			}
1683 
1684 			/*
1685 			 * Slow start out of the error condition.  We
1686 			 * use the MTU because we know it's smaller
1687 			 * than the previously transmitted segment.
1688 			 *
1689 			 * Note: This is more conservative than the
1690 			 * suggestion in draft-floyd-incr-init-win-03.
1691 			 */
1692 			if (rt->rt_rmx.rmx_mtu != 0)
1693 				tp->snd_cwnd =
1694 				    TCP_INITIAL_WINDOW(tcp_init_win,
1695 				    rt->rt_rmx.rmx_mtu);
1696 		}
1697 
1698 		/*
1699 		 * Resend unacknowledged packets.
1700 		 */
1701 		tp->snd_nxt = tp->snd_una;
1702 		tcp_output(tp);
1703 	}
1704 }
1705 #endif /* INET6 */
1706 
1707 /*
1708  * Compute the MSS to advertise to the peer.  Called only during
1709  * the 3-way handshake.  If we are the server (peer initiated
1710  * connection), we are called with a pointer to the interface
1711  * on which the SYN packet arrived.  If we are the client (we
1712  * initiated connection), we are called with a pointer to the
1713  * interface out which this connection should go.
1714  *
1715  * NOTE: Do not subtract IP option/extension header size nor IPsec
1716  * header size from MSS advertisement.  MSS option must hold the maximum
1717  * segment size we can accept, so it must always be:
1718  *	 max(if mtu) - ip header - tcp header
1719  */
1720 u_long
1721 tcp_mss_to_advertise(ifp, af)
1722 	const struct ifnet *ifp;
1723 	int af;
1724 {
1725 	extern u_long in_maxmtu;
1726 	u_long mss = 0;
1727 	u_long hdrsiz;
1728 
1729 	/*
1730 	 * In order to avoid defeating path MTU discovery on the peer,
1731 	 * we advertise the max MTU of all attached networks as our MSS,
1732 	 * per RFC 1191, section 3.1.
1733 	 *
1734 	 * We provide the option to advertise just the MTU of
1735 	 * the interface on which we hope this connection will
1736 	 * be receiving.  If we are responding to a SYN, we
1737 	 * will have a pretty good idea about this, but when
1738 	 * initiating a connection there is a bit more doubt.
1739 	 *
1740 	 * We also need to ensure that loopback has a large enough
1741 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1742 	 */
1743 
1744 	if (ifp != NULL)
1745 		switch (af) {
1746 		case AF_INET:
1747 			mss = ifp->if_mtu;
1748 			break;
1749 #ifdef INET6
1750 		case AF_INET6:
1751 			mss = IN6_LINKMTU(ifp);
1752 			break;
1753 #endif
1754 		}
1755 
1756 	if (tcp_mss_ifmtu == 0)
1757 		switch (af) {
1758 		case AF_INET:
1759 			mss = max(in_maxmtu, mss);
1760 			break;
1761 #ifdef INET6
1762 		case AF_INET6:
1763 			mss = max(in6_maxmtu, mss);
1764 			break;
1765 #endif
1766 		}
1767 
1768 	switch (af) {
1769 	case AF_INET:
1770 		hdrsiz = sizeof(struct ip);
1771 		break;
1772 #ifdef INET6
1773 	case AF_INET6:
1774 		hdrsiz = sizeof(struct ip6_hdr);
1775 		break;
1776 #endif
1777 	default:
1778 		hdrsiz = 0;
1779 		break;
1780 	}
1781 	hdrsiz += sizeof(struct tcphdr);
1782 	if (mss > hdrsiz)
1783 		mss -= hdrsiz;
1784 
1785 	mss = max(tcp_mssdflt, mss);
1786 	return (mss);
1787 }
1788 
1789 /*
1790  * Set connection variables based on the peer's advertised MSS.
1791  * We are passed the TCPCB for the actual connection.  If we
1792  * are the server, we are called by the compressed state engine
1793  * when the 3-way handshake is complete.  If we are the client,
1794  * we are called when we receive the SYN,ACK from the server.
1795  *
1796  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1797  * before this routine is called!
1798  */
1799 void
1800 tcp_mss_from_peer(tp, offer)
1801 	struct tcpcb *tp;
1802 	int offer;
1803 {
1804 	struct socket *so;
1805 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1806 	struct rtentry *rt;
1807 #endif
1808 	u_long bufsize;
1809 	int mss;
1810 
1811 #ifdef DIAGNOSTIC
1812 	if (tp->t_inpcb && tp->t_in6pcb)
1813 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1814 #endif
1815 	so = NULL;
1816 	rt = NULL;
1817 #ifdef INET
1818 	if (tp->t_inpcb) {
1819 		so = tp->t_inpcb->inp_socket;
1820 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1821 		rt = in_pcbrtentry(tp->t_inpcb);
1822 #endif
1823 	}
1824 #endif
1825 #ifdef INET6
1826 	if (tp->t_in6pcb) {
1827 		so = tp->t_in6pcb->in6p_socket;
1828 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1829 		rt = in6_pcbrtentry(tp->t_in6pcb);
1830 #endif
1831 	}
1832 #endif
1833 
1834 	/*
1835 	 * As per RFC1122, use the default MSS value, unless they
1836 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1837 	 */
1838 	mss = tcp_mssdflt;
1839 	if (offer)
1840 		mss = offer;
1841 	mss = max(mss, 32);		/* sanity */
1842 	tp->t_peermss = mss;
1843 	mss -= tcp_optlen(tp);
1844 #ifdef INET
1845 	if (tp->t_inpcb)
1846 		mss -= ip_optlen(tp->t_inpcb);
1847 #endif
1848 #ifdef INET6
1849 	if (tp->t_in6pcb)
1850 		mss -= ip6_optlen(tp->t_in6pcb);
1851 #endif
1852 
1853 	/*
1854 	 * If there's a pipesize, change the socket buffer to that size.
1855 	 * Make the socket buffer an integral number of MSS units.  If
1856 	 * the MSS is larger than the socket buffer, artificially decrease
1857 	 * the MSS.
1858 	 */
1859 #ifdef RTV_SPIPE
1860 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1861 		bufsize = rt->rt_rmx.rmx_sendpipe;
1862 	else
1863 #endif
1864 		bufsize = so->so_snd.sb_hiwat;
1865 	if (bufsize < mss)
1866 		mss = bufsize;
1867 	else {
1868 		bufsize = roundup(bufsize, mss);
1869 		if (bufsize > sb_max)
1870 			bufsize = sb_max;
1871 		(void) sbreserve(&so->so_snd, bufsize);
1872 	}
1873 	tp->t_segsz = mss;
1874 
1875 #ifdef RTV_SSTHRESH
1876 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1877 		/*
1878 		 * There's some sort of gateway or interface buffer
1879 		 * limit on the path.  Use this to set the slow
1880 		 * start threshold, but set the threshold to no less
1881 		 * than 2 * MSS.
1882 		 */
1883 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1884 	}
1885 #endif
1886 }
1887 
1888 /*
1889  * Processing necessary when a TCP connection is established.
1890  */
1891 void
1892 tcp_established(tp)
1893 	struct tcpcb *tp;
1894 {
1895 	struct socket *so;
1896 #ifdef RTV_RPIPE
1897 	struct rtentry *rt;
1898 #endif
1899 	u_long bufsize;
1900 
1901 #ifdef DIAGNOSTIC
1902 	if (tp->t_inpcb && tp->t_in6pcb)
1903 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1904 #endif
1905 	so = NULL;
1906 	rt = NULL;
1907 #ifdef INET
1908 	if (tp->t_inpcb) {
1909 		so = tp->t_inpcb->inp_socket;
1910 #if defined(RTV_RPIPE)
1911 		rt = in_pcbrtentry(tp->t_inpcb);
1912 #endif
1913 	}
1914 #endif
1915 #ifdef INET6
1916 	if (tp->t_in6pcb) {
1917 		so = tp->t_in6pcb->in6p_socket;
1918 #if defined(RTV_RPIPE)
1919 		rt = in6_pcbrtentry(tp->t_in6pcb);
1920 #endif
1921 	}
1922 #endif
1923 
1924 	tp->t_state = TCPS_ESTABLISHED;
1925 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1926 
1927 #ifdef RTV_RPIPE
1928 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1929 		bufsize = rt->rt_rmx.rmx_recvpipe;
1930 	else
1931 #endif
1932 		bufsize = so->so_rcv.sb_hiwat;
1933 	if (bufsize > tp->t_ourmss) {
1934 		bufsize = roundup(bufsize, tp->t_ourmss);
1935 		if (bufsize > sb_max)
1936 			bufsize = sb_max;
1937 		(void) sbreserve(&so->so_rcv, bufsize);
1938 	}
1939 }
1940 
1941 /*
1942  * Check if there's an initial rtt or rttvar.  Convert from the
1943  * route-table units to scaled multiples of the slow timeout timer.
1944  * Called only during the 3-way handshake.
1945  */
1946 void
1947 tcp_rmx_rtt(tp)
1948 	struct tcpcb *tp;
1949 {
1950 #ifdef RTV_RTT
1951 	struct rtentry *rt = NULL;
1952 	int rtt;
1953 
1954 #ifdef DIAGNOSTIC
1955 	if (tp->t_inpcb && tp->t_in6pcb)
1956 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1957 #endif
1958 #ifdef INET
1959 	if (tp->t_inpcb)
1960 		rt = in_pcbrtentry(tp->t_inpcb);
1961 #endif
1962 #ifdef INET6
1963 	if (tp->t_in6pcb)
1964 		rt = in6_pcbrtentry(tp->t_in6pcb);
1965 #endif
1966 	if (rt == NULL)
1967 		return;
1968 
1969 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1970 		/*
1971 		 * XXX The lock bit for MTU indicates that the value
1972 		 * is also a minimum value; this is subject to time.
1973 		 */
1974 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1975 			TCPT_RANGESET(tp->t_rttmin,
1976 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1977 			    TCPTV_MIN, TCPTV_REXMTMAX);
1978 		tp->t_srtt = rtt /
1979 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1980 		if (rt->rt_rmx.rmx_rttvar) {
1981 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1982 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1983 				(TCP_RTTVAR_SHIFT + 2));
1984 		} else {
1985 			/* Default variation is +- 1 rtt */
1986 			tp->t_rttvar =
1987 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1988 		}
1989 		TCPT_RANGESET(tp->t_rxtcur,
1990 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1991 		    tp->t_rttmin, TCPTV_REXMTMAX);
1992 	}
1993 #endif
1994 }
1995 
1996 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1997 #if NRND > 0
1998 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1999 #endif
2000 
2001 /*
2002  * Get a new sequence value given a tcp control block
2003  */
2004 tcp_seq
2005 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2006 {
2007 
2008 #ifdef INET
2009 	if (tp->t_inpcb != NULL) {
2010 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2011 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2012 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2013 		    addin));
2014 	}
2015 #endif
2016 #ifdef INET6
2017 	if (tp->t_in6pcb != NULL) {
2018 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2019 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2020 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2021 		    addin));
2022 	}
2023 #endif
2024 	/* Not possible. */
2025 	panic("tcp_new_iss");
2026 }
2027 
2028 /*
2029  * This routine actually generates a new TCP initial sequence number.
2030  */
2031 tcp_seq
2032 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2033     size_t addrsz, tcp_seq addin)
2034 {
2035 	tcp_seq tcp_iss;
2036 
2037 #if NRND > 0
2038 	static int beenhere;
2039 
2040 	/*
2041 	 * If we haven't been here before, initialize our cryptographic
2042 	 * hash secret.
2043 	 */
2044 	if (beenhere == 0) {
2045 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2046 		    RND_EXTRACT_ANY);
2047 		beenhere = 1;
2048 	}
2049 
2050 	if (tcp_do_rfc1948) {
2051 		MD5_CTX ctx;
2052 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2053 
2054 		/*
2055 		 * Compute the base value of the ISS.  It is a hash
2056 		 * of (saddr, sport, daddr, dport, secret).
2057 		 */
2058 		MD5Init(&ctx);
2059 
2060 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2061 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2062 
2063 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2064 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2065 
2066 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2067 
2068 		MD5Final(hash, &ctx);
2069 
2070 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2071 
2072 		/*
2073 		 * Now increment our "timer", and add it in to
2074 		 * the computed value.
2075 		 *
2076 		 * XXX Use `addin'?
2077 		 * XXX TCP_ISSINCR too large to use?
2078 		 */
2079 		tcp_iss_seq += TCP_ISSINCR;
2080 #ifdef TCPISS_DEBUG
2081 		printf("ISS hash 0x%08x, ", tcp_iss);
2082 #endif
2083 		tcp_iss += tcp_iss_seq + addin;
2084 #ifdef TCPISS_DEBUG
2085 		printf("new ISS 0x%08x\n", tcp_iss);
2086 #endif
2087 	} else
2088 #endif /* NRND > 0 */
2089 	{
2090 		/*
2091 		 * Randomize.
2092 		 */
2093 #if NRND > 0
2094 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2095 #else
2096 		tcp_iss = arc4random();
2097 #endif
2098 
2099 		/*
2100 		 * If we were asked to add some amount to a known value,
2101 		 * we will take a random value obtained above, mask off
2102 		 * the upper bits, and add in the known value.  We also
2103 		 * add in a constant to ensure that we are at least a
2104 		 * certain distance from the original value.
2105 		 *
2106 		 * This is used when an old connection is in timed wait
2107 		 * and we have a new one coming in, for instance.
2108 		 */
2109 		if (addin != 0) {
2110 #ifdef TCPISS_DEBUG
2111 			printf("Random %08x, ", tcp_iss);
2112 #endif
2113 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2114 			tcp_iss += addin + TCP_ISSINCR;
2115 #ifdef TCPISS_DEBUG
2116 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2117 #endif
2118 		} else {
2119 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2120 			tcp_iss += tcp_iss_seq;
2121 			tcp_iss_seq += TCP_ISSINCR;
2122 #ifdef TCPISS_DEBUG
2123 			printf("ISS %08x\n", tcp_iss);
2124 #endif
2125 		}
2126 	}
2127 
2128 	if (tcp_compat_42) {
2129 		/*
2130 		 * Limit it to the positive range for really old TCP
2131 		 * implementations.
2132 		 * Just AND off the top bit instead of checking if
2133 		 * is set first - saves a branch 50% of the time.
2134 		 */
2135 		tcp_iss &= 0x7fffffff;		/* XXX */
2136 	}
2137 
2138 	return (tcp_iss);
2139 }
2140 
2141 #ifdef IPSEC
2142 /* compute ESP/AH header size for TCP, including outer IP header. */
2143 size_t
2144 ipsec4_hdrsiz_tcp(tp)
2145 	struct tcpcb *tp;
2146 {
2147 	struct inpcb *inp;
2148 	size_t hdrsiz;
2149 
2150 	/* XXX mapped addr case (tp->t_in6pcb) */
2151 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2152 		return 0;
2153 	switch (tp->t_family) {
2154 	case AF_INET:
2155 		/* XXX: should use currect direction. */
2156 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2157 		break;
2158 	default:
2159 		hdrsiz = 0;
2160 		break;
2161 	}
2162 
2163 	return hdrsiz;
2164 }
2165 
2166 #ifdef INET6
2167 size_t
2168 ipsec6_hdrsiz_tcp(tp)
2169 	struct tcpcb *tp;
2170 {
2171 	struct in6pcb *in6p;
2172 	size_t hdrsiz;
2173 
2174 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2175 		return 0;
2176 	switch (tp->t_family) {
2177 	case AF_INET6:
2178 		/* XXX: should use currect direction. */
2179 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2180 		break;
2181 	case AF_INET:
2182 		/* mapped address case - tricky */
2183 	default:
2184 		hdrsiz = 0;
2185 		break;
2186 	}
2187 
2188 	return hdrsiz;
2189 }
2190 #endif
2191 #endif /*IPSEC*/
2192 
2193 /*
2194  * Determine the length of the TCP options for this connection.
2195  *
2196  * XXX:  What do we do for SACK, when we add that?  Just reserve
2197  *       all of the space?  Otherwise we can't exactly be incrementing
2198  *       cwnd by an amount that varies depending on the amount we last
2199  *       had to SACK!
2200  */
2201 
2202 u_int
2203 tcp_optlen(tp)
2204 	struct tcpcb *tp;
2205 {
2206 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2207 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2208 		return TCPOLEN_TSTAMP_APPA;
2209 	else
2210 		return 0;
2211 }
2212